1 / 8
Data redaction patterns + safe-prompt construction
In this lesson
Data redaction patterns + safe-prompt construction
Configure AWS Bedrock Guardrails to block or anonymize personally identifiable information (PII) in both model inputs and outputs, selecting appropriate action modes (BLOCK, ANONYMIZE, or NONE) for at least five…
You'll be able to
- Configure AWS Bedrock Guardrails to block or anonymize personally identifiable information (PII) in both model inputs and outputs, selecting appropriate action modes (BLOCK, ANONYMIZE, or NONE) for at least five built-in PII entity types[^1][^3].
- Apply automated PII detection and redaction techniques to production ML pipelines, implementing solutions such as Amazon Comprehend for text data and Amazon Transcribe's automatic redaction for audio transcriptions[^4].
- Design prompt templates that minimize sensitive data exposure by replacing direct PII with placeholder variables and XML-structured content delimiters, ensuring prompts summarize or paraphrase rather than paste verbatim sensitive information[^2][^5].
- Evaluate guardrail policy configurations using detect mode to identify false positives and negatives before deploying content filters and sensitive information policies to production environments[^6].
- Implement custom regular expression patterns within guardrails to detect and mask organization-specific sensitive data types not covered by built-in PII entities, configuring separate input and output actions as required by your use case[^3].